Characterization of bone abnormalities from micro-CT images for evaluating drug toxicity in developmental and reproductive toxicology (DART) studies

Abstract

Routinely, compounds are assessed by developmental and reproductive toxicology (DART) studies to evaluate the potential for drug-induced birth defects. High-throughput micro-CT images are being used to evaluate skeletal abnormalities due to its ability to provide high quality images of bone structures. Currently, these micro-CT images are visually inspected for skeletal abnormalities, which is a time and resource intensive process. To reduce the resources needed for skeletal evaluation, we developed image analysis strategies that allow for automatic segmentation of whole body CT images into individual bones and use structural variations of shape characteristics to classify bones as normal or abnormal. Extraction of various structures in the skull and torso were accomplished sequentially starting with skull bones and moving towards the neck, vertebrae, ribs, and limbs. A total of 17 skull bones/structures (supraoccipital, mandible, squamosals, zygomatics, etc.) and 20 torso structures (ribs, spine, humerus, femur, tibia, etc.) were identified and isolated using this algorithm. Next, we used geometrical (volume, length, width, etc.) and shape-based characteristics to identify bones lying outside the normal distribution of numbers, shapes and sizes to flag fetuses for potential abnormalities. We applied this tool to a test data set of 167 fetuses with verified skeletal abnormalities and received sensitivity of 0.959 and specificity of 0.805. This analysis platform allows for fully automated batch processing of images. Future work will include further development of the current platform to improve performance.